Regional Diffusion of Divorce in Turkey

28 Pages·2017·0.94 MB·English

by  Kim Caarls

EurJPopulation(2018)34:609–636
https://doi.org/10.1007/s10680-017-9441-5
Regional Diffusion of Divorce in Turkey
Kim Caarls1 · Helga A. G. de Valk1
Received:5July2016/Accepted:2August2017/Publishedonline:3October2017
©TheAuthor(s)2017.Thisarticleisanopenaccesspublication
Abstract WhiledemographicchangehasbeenwelldocumentedformanyWestern
countries, much less is known about demographic transitions in other countries,
including Turkey. Demographic change in European societies can be characterized
by,amongstothers,increasedprevalenceofdivorce.Althoughitisoftenarguedthat
lifecoursesinTurkeyfollowamoretraditionalpath,littleisknownondeterminants
and patterns of divorce, despite the major socioeconomic changes Turkey has
undergone over the past decades. We study the levels of divorce of women in
Turkey from 1973 to 2008 to explain patterns of divorce, looking at the role of
individualcharacteristicsandtheregionalcontext.WeusetheDemographicHealth
Surveys (2003/2008), complemented with regional data on divorce, urbanization,
andGDPpercapita.Applyingamultilevelapproach,distinguishing12regions,we
hypothesize that regions where divorce is already more prevalent, more urbanized
regions, and wealthier regions in terms of GDP per capita will increase the prob-
ability of divorce. Our analyses show that levels of divorce increased over the past
decades but huge regional variation remains. Sociocultural and socioeconomic
factorsexplainthistrend,andinparticularurbanizationandGDPpercapitaarekey
determinants for divorce.
Keywords Divorce·Turkey·Regionalcontext·Multilevel·Eventhistoryanalysis
& KimCaarls
[email protected]
HelgaA.G.deValk
[email protected]
1 NetherlandsInterdisciplinaryDemographicInstitute(NIDI)/KoninklijkeNederlandse
AkademievanWetenschappen(KNAW)/UniversityofGroningen(UG),LangeHoutstraat19,
2511CVTheHague,TheNetherlands
123
610 K.Caarls,H.A.G.deValk
1 Introduction
Divorcehasbeenthetopicofextensiveresearchduringthepastdecades(foroverview
articlesseee.g.,Amato2010;AmatoandJames2010;Ha¨rko¨nen2013;Lyngstadand
Jalovaara 2010; Wagner and Weiβ 2006). Increasing levels of divorce and other
demographic changes have been linked to a shift in ideas toward family life where
individualisticattitudesandhigheracceptanceofdivergentfamilybehavioursprevail
(e.g. Ha¨rko¨nen 2013; Lesthaeghe 1995). While these processes have been well
documentedinWesterncountries,andinparticulartheUSAandEurope,muchlessis
knownaboutdemographictransitionsinotherregionsoftheworld,includingTurkey
(Adams2004;Rashad2000;TabutinandSchoumaker2005).
It is often argued that life courses in Turkey still follow a more traditional path,
yet we know little about family life transitions in general and the patterns and
determinants of divorce in Turkey in particular. While higher divorce rates in
Europereflect,amongstothers,thechangeddemographicbehaviour(e.g.Ha¨rko¨nen
2013;Lesthaeghe1983,2010),itisoftenarguedthatthe demographictransitionin
Turkey is not as advanced (Rashad 2000; Yu¨c¸es¸ahin and O¨zgu¨r 2008). However,
Turkeyhasundergonemajorsocioeconomicchangesoverthepastdecadesandthis
had its impact on family life transitions (Yu¨c¸es¸ahin and O¨zgu¨r 2008; Yu¨ksel-
KaptanogluandErgo¨c¸men2014).Turkeyhasnotonlywitnessedanotableincrease
in divorce rates during the past decades (e.g. Demir 2013; Ha¨rko¨nen 2013; Kavas
andGu¨ndu¨z-Hoșgo¨r2010;TurkishStatisticalInstitute(TurkStat)2011),thisgrowth
inlevelsofdivorcecorrespondstochangesinTurkishfamilylifeinmanydomains
duringthelastcentury:peoplemarrylaterandhavefewerchildren,andgenderroles
are said to be less traditional (Kavas and Thornton 2013).
These changes have been attributed to modernization processes, exposure to
Western values, and socioeconomic changes. In addition, and reflecting these
changes, there were several legislative developments in Turkey that affected the
family.Regardingdivorce,thetwomajorchangesweretheintroductionofthe1988
no-fault divorce law and the amendments to the Turkish Civil Code in 2001, both
expandingwomen’srightsandadvancingtheirpositioninthemarriage(Arat2010;
Kavas and Gu¨ndu¨z-Hoșgo¨r 2010; Yu¨ksel-Kaptanoglu and Ergo¨c¸men 2014).
Moreover,whatalsomakesTurkeyaninterestingcaseisthehugeregionalvariation:
therearesubstantialdifferencesbetweenregionsbothintermsofeconomicdevelopment
as well as in the spread of more modern values toward family life. Considering total
fertility rates (TFR), for example, in some regions, these rates proximate those of
Europeancountries,whileinotherregionstheTFRhasremainedhigh(Yu¨c¸es¸ahinand
O¨zgu¨r 2008; Yavuz 2008). While most studies on divorce concentrate on individual
socioeconomicanddemographicfactorsthatpredictwhetheracoupledivorcesornot(e.
g.Heaton2002;WagnerandWeiβ2006),macro-levelfactorsshapethecontextinwhich
acouple’sunionformationanddissolutiontakesplace.Particularly,theroleofregional
variationwithinonecountryhasremainedrelativelyunderstudied(butsee,e.g.Kalmijn
andUnnk2007;Lester1999;GlassandLevchak2014).
The aim of our paper is twofold. First,we examinedivorce patterns in Turkey’s
12regionsovera40-yeartimeperiod,between1967and2008.Westudymarriage
123
RegionalDiffusionofDivorceinTurkey 611
cohort and period effects (e.g. the impact of different laws) amongst women aged
15–49 years. Second, we aim to pinpoint the mechanisms explaining the regional
variation of divorce inTurkey, by investigatingthe importance of both macro- and
micro-levelpredictorssimultaneously.Atthemacro-level,botheconomicfactorsas
wellasthespreadofmodernvaluesmayinfluencetheprobabilityofdivorce.More
specifically, we investigate the role of regional variation on the probability of
divorce by considering the average share of the regional gross domestic product
(GDP) per capita within the total country-level GDP per capita, the level of
populationdensity,andcrudedivorcerates(CDR)foreachofTurkey’s12regions.
Data come from the Demographic Health Survey (DHS) (2003 and 2008 waves,
http://www.dhsprogram.com)enrichedwithregionaldatafromTurkStat.Multilevel
discrete-time event history models are used to examine to what degree individual
characteristics and the regional context influence divorce behaviour.
2 Divorce in Turkey
In Western countries, divorce levels began rising sharply from the 1950 s onward
(Lesthaeghe2010).InTurkey,anincreaseindivorceratesoccurredlater(Ha¨rko¨nen
2013).Yetoncedivorcerateswererising,changesweresubstantial:whiletheCDR
wasonly0.27in1970,itincreasedto1.40in2008(TurkStat2011).Paralleltothis
growthinlevelsofdivorceareotherchangesinTurkishfamilylife.Inadditiontoa
rising prevalence of divorce, fertility and mortality have been decreasing, bringing
Turkey’s reproduction close to replacement level (DHS 2009). Furthermore, age at
marriage increased, gender roles became more egalitarian, and the prevalence of
patriarchalextendedfamiliesdeclined(Yu¨ksel-KaptanogluandErgo¨c¸men2014).In
the same vein, attitudes toward divorce became more tolerant (e.g. Kavas and
Thornton 2013; TurkStat 2011). Although these changes have been connected to
modernization processes and exposure to Western values, local values are not
necessarilyabandonedandmodernandtraditionalvaluesaresimultaneouslypresent
in Turkish society (Kavas and Gu¨ndu¨z-Hoșgo¨r 2010; Kavas and Thornton 2013;
Yu¨c¸es¸ahin and O¨zgu¨r 2008).
Although Turkish society can be characterized as patriarchal with low levels of
gender equality (Go¨ksel 2013; United Nations Development Programme (UNDP)
2005), progress has been made in this area (Yu¨ksel-Kaptanoglu and Ergo¨c¸men
2014). Women in Turkey increasingly challenge existing gender norms. For
example, while financial decision-making was traditionally an exclusively male
affairandjointbankaccountswerebasicallynon-existent,Turkishwomentodayare
managing their own bank accounts to a greater extent (Kavas and Gu¨ndu¨z-Hoșgo¨r
2010). Despite these changes in gender relations that took place in Turkish family
life, divorced women are still stigmatized and held accountable for their broken
marriage (Kavas and Gu¨ndu¨z-Hoșgo¨r 2011; O¨zar and Yakut-Cakar 2013).
Withrespecttodivorce,therehavebeenseveralrelevantchangesinTurkishlaw.
ShortlyafterthefoundationoftheTurkishRepublicin1923,theTurkishCivilCode
was adopted in 1926. This code provided women with more progressive rights
comparedtotheSharialawthatwaspracticedbefore,suchasequalinheritanceand
123
612 K.Caarls,H.A.G.deValk
divorce rights for men and women. Although this code provided more gender
equality, it was not until the 1980s that critiques were voiced against this code,
whichstillreflectedstrongpatriarchalnotionsregardingfamilymatters(Arat2010).
In particular, articles underpinning women’s subordinate position were criticized,
such as those defining the husband as head of the family and his wife as his helper
and those formalizing separate ownership in marriage. These articles were
especially detrimental for women during divorce, as they were consequently left
without income or property.
Due to the efforts of the women’s movement (amongst others), some minor
reforms took place before 2001. Turkey has signed several international conven-
tions, such as the Convention on the Elimination of All Forms of Discrimination
against Women monitored by the Committee on the Elimination of Discrimination
against Women (Yu¨ksel-Kaptanoglu and Ergo¨c¸men 2014). Most notably, the no-
fault divorce was introduced in 1988 [Divorce Law (No. 3444)], allowing divorce
by mutual consent (Kavas and Gu¨ndu¨z-Hoșgo¨r 2010).
Yet the most significant changes took place with the 2001 amendments of the
TurkishCivilCode,whichsignificantlyexpandedwomen’srights,inparticularwith
respecttowomen’spositioninthemarriage:articlesthatdeclaredthehusbandtobe
theheadofthehouseholdandhiswifeashishelperweredeleted,theminimumageof
marriage was raised to 18 for both men and women (it used to be 17 and 15,
respectively), there were major changes in the property regime, from one based on
separateownershiptoonebasedonthesharingofproperty.Thisimpliedthatinthe
eventofdivorce,womencouldnowclaimashareofthepropertyregisteredintheir
husbandsnameifthepropertywasacquiredduringthemarriage(seeArat2010;Kavas
and Gu¨ndu¨z-Hoșgo¨r 2010). Furthermore, while fathers originally had the final say
overchildcustody,thenewcodeprovidedfathersandmotherswithequalleverage.
These changes in legislation took place in a context of socioeconomic
developments, such as industrialization, rapid urbanization, educational expansion
[particularly for women, e.g. the share of women with secondary education has
increased from 34% in 1997 to 64% in 2010 (UNDP 2013)], and increased
participation of women in economic spheres. These developments had their
repercussions on Turkish family life, but these developments have not been
distributedequallyacrossTurkey.Modernandtraditionalvaluesaresimultaneously
present in Turkish society (Cindoglu et al. 2008; Kavas and Gu¨ndu¨z-Hoșgo¨r 2010;
Kavas and Thornton 2013), resulting in diverse family systems across regions.
While family regulations changed, practices such as arranged marriages, consan-
guineousmarriages,andreligiousmarriagesmaintained,insomeregionsmorethan
others (Kavas 2010; Saadat 2015).
3 Regional Variation in Divorce
In2002,TurkeywasincludedintheNomenclatureofTerritorialUnitsforStatistics
(NUTS) within the framework of the EU harmonization process, and 12 distinct
regions (NUTS I) were identified. To a large extent, the sociocultural, sociodemo-
graphic and socioeconomic differences within Turkey are mirrored in these 12
123
RegionalDiffusionofDivorceinTurkey 613
regions (DHS 2009). The regions in the Western part of Turkey, encompassing
Istanbul and I˙zmir, are the most urbanized and industrialized. The regions in the
Southhaveseveralgrowingindustrialcentres,suchasAdana,Mersin,andAntalya.
Thecapitalcity,Ankara,liesinCentralTurkey.Besidesthismetropolis,theregions
in Central Turkey are moderately industrial. The Northern region has a fertile
coastal line and a mountainous interior, mainly occupied by small-scale farmers.
TheEastern regions are economicallythe least developed andcan be characterized
by a rugged landscape.
WhileTurkey’sCDR(thenumberofdivorcesperthousandpopulationinagiven
year) has been relatively low in recent years, the prevalence of divorce differs
greatlybetweenthedifferentregions.Whereastheoveralldivorceratein2008was
1.40, it ranged from 0.48 in Southeast Anatolia to 1.88 in the Aegean region
(TurkStat 2011), which for example equals the crude divorce rates of the
Netherlands [1.9 in 2009 (Eurostat 2015)]. Several macro-level studies indicated
a range ofcontextual factors that are correlated with the cross-national variationin
divorce rates, such as the role of the normative context (Amato and Keith 1991;
Lesthaeghe 1995; Wagner and Weiβ 2006; Wolfinger 1999), legislative changes
toward more liberal divorce legislation (Gonzale´z and Viitanen 2009; Stevenson
and Wolfers 2007; Wolfinger 1999), family policies (Engelhardt et al. 2002), and
femalelabourmarketparticipation(DiekmannandSchmidheiny2004;Kalmijnand
Unnk 2007). While these studies typically analyse between-country variation, we
are interested in variation between regions. These regions can be considered a
relevant context as they provide local opportunity structures (e.g. degree of
urbanization, socioeconomic situation) and cultural milieus (e.g. acceptance or
prevalence of divorce) that can affect individual behaviour (Hank 2002).
We expect that not only socioeconomic features of these regions play a major
role in the level of diffusion of divorce (over and beyond the role of individual
socioeconomic and sociodemographic characteristics), but also that different
sociocultural factors influence the prevalence of divorce. For example, social
norms regarding the use and acceptance of birth control vary strongly in the
different regions in Turkey (Yu¨c¸es¸ahin and O¨zgu¨r 2008). Previous studies on
fertility in Turkey demonstrated that these different reproductive behaviours of
women could be explained by, amongst others, diffusion processes (Yavuz 2008;
Yu¨c¸es¸ahin and O¨zgu¨r 2008). In particular, the cultural isolation of the eastern
regions has prevented the diffusion of new and innovative reproductive behaviour,
resultinginhighfertilityratesintheeasternpartcontrarytodecliningfertilitytrends
elsewhere in Turkey (Yu¨c¸es¸ahin and O¨zgu¨r 2008).
Inasimilarvein,weexpectsocialnormsregardingdivorcetodifferbetweenthe
12regions.Consequently,thesedifferentnormswillresultindifferentCDRinthese
regions,sinceahigherprevalenceofdivorcereflectsahigherculturalacceptanceof
divorce (Ha¨rko¨nen 2013). It can be expected that higher acceptance will decrease
the stigmatization of divorce, making divorce more accessible for women. We
therefore hypothesize that in regions where divorce has been more prevalent in the
past, women’s probability of divorce in later years will be higher compared to
regionswithlowerlevelsofdivorce,netofotherregional-levelandindividual-level
characteristics (Hypothesis 1).
123
614 K.Caarls,H.A.G.deValk
Accordingtothediffusiontheory,newbehaviourstypicallystartinmetropolitan
areas, where the upper and middle classes take the lead (Liefbroer and Doureleijn
2006;NazioandBlossfeld2002;Reedetal.1999;Rogers1983;StrangandMeyer
1993;StrangandTuma1993;Yavuz2008).Thelevelofurbanizationvariesgreatly
betweenthe12regions,andweexpectthelevelofdivorcetovaryaccordingly,with
a higher probability of divorce for women that live in more urbanized regions
compared to women that live in intermediate or rural regions (Hypothesis 2).
There is also huge regional variation in terms of socioeconomic development.
RegionsintheWesternparthaveamuchlargershareofthenationalgrossdomestic
product (GDP) per capita than the Northern or Eastern regions (DHS 2009).
Existing micro-level studies on divorce show mixed findings when it comes to the
effect of economic circumstances on the probability of divorce (Aytac¸ and Rankin
2009;Ha¨rko¨nenandDronkers2006;Jalovaara2003;LyngstadandJalovaara2010).
Several scholars studying these micro-level effects argue that acquiring a higher
income has a stabilizing effect on marriages (the “income effect”). However, there
isampleevidencefromEuropeandNorthAmericathatthis“incomeeffect”willbe
outweighed by the so-called “independence effect”, which refers to an increase in
female participation on the labour market resulting in more instable marriages
(LyngstadandJalovaara2010).ConsideringtheTurkishcontext,weexpectthatthe
“independence effect” will be minimal, as female labour market participation is
extremely low (ILO 2016). Yet few studies have analysed the impact of
macroeconomic circumstances on divorce (for exceptions see, e.g. Amato and
Beattie2011;BremmerandKesselring2004;Schaller2012),andevenfewerlooked
attheeffectofmacroeconomicconditionsontheprobabilityofdivorceonamicro-
level (e.g. Fischer and Liefbroer 2006). As with micro-level studies, previous
macro-level studiestypicallydiscuss two opposinghypotheses. On the onehand, it
is suggested that economic hardship increases the chance of divorce, with worse
economic circumstances leading to more divorce and better economic conditions
resultinginlessdivorce.Ontheotherhand,difficulteconomicconditionsmaymake
itchallengingtocoverthecostsofdivorce,makingdivorcelesslikely(Fischerand
Liefbroer 2006). To account for the specific Turkish context, where the social and
financial cost for divorce is high, we expect the second hypothesis to hold: bad
economicconditionswillleadtolessdivorce(Aytac¸ andRankin2009;Fischerand
Liefbroer 2006; Kavas and Gu¨ndu¨z-Hoșgo¨r 2010) (Hypothesis 3).
4 Changes Over Time: Cohort and Period Effects
Divorces in Turkey have significantly increased over the past 40 years, with CDR
rising from0.27in1970to1.40in2008(TurkStat2011).Inadditiontoexamining
the role of individual and regional characteristics, we are also interested in
explaining this increase in divorce over time, and to investigate whether period or
cohort effects drive this change. Marriage cohort effects relate to the timing of
marriage and the conditions that were present at that time. Consequently, different
marriage cohorts have different attitudes, resources and practices, and these
differences affect divorce rates (Ha¨rko¨nen 2013; Lyngstad and Jalovaara 2010).
123
RegionalDiffusionofDivorceinTurkey 615
Period effects, in turn, affect all married couples, regardless of when they were
married. The impact of new laws, for example,can bring about such period effects
(Gonza´lez and Viitanen 2009; Yu¨ksel-Kaptanoglu and Ergo¨c¸men 2014).
Weexploredbothcohortandperiodeffectstoinvestigate whetherthechangein
CDRovertimeistheresultoftheemancipatingeffectoftheno-faultdivorcelawin
1988 or of the amendments in Turkey’s Civil Code in 2001 (period effects) or
whether these changes took place due to the changing social context wherein
marriages took place (cohort effects). If period effects are present, significant
changes should be seen between the periods after the 1988 and 2001 reforms
(Hypothesis4a).Ifcohorteffectsexplainthechangesovertime,weshouldseemore
gradual changes over the studied time span (Hypothesis 4b).
Previous studies also addressed the issue whether the predictors of divorce
changeovertime(BernardiandMart´ınez-Pastor2011;DeGraafandKalmijn2006;
Ha¨rko¨nenandDronkers2006;Ha¨rko¨nen2013).Inparticular,previousstudieshave
shown the effect ofwomen’seducationon divorce tochange over time.According
to the so-called Goode hypothesis, the society presents a normative context that
shapes individual divorce behaviour; when, in a given context, divorce is a
relatively rare and often stigmatized event, it takes more resources to dissolve a
marriage (Goode 1962). This implies that when divorce is not so common, higher
educated women are more likely to break up (De Graaf and Kalmijn 2006;
Ha¨rko¨nen and Dronkers 2006; Bernardi and Mart´ınez-Pastor 2011). Additionally,
womenwithhighersocioeconomicstatuswillbemostlikelytobetheearlyadapters
or innovators of new behaviours, such as divorce (Blossfeld et al. 1995; Hoem
1997).
For the Turkish fertility transition, higher educated women speaking Turkish
were identified as the pioneers (Yavuz 2008). A higher socioeconomic status also
makes women less sensitive to social conformities, even in spite of a sociocultural
contextinTurkeythattypicallyrepresentspatriarchalnormsandvalues(Kavasand
Gu¨ndu¨z-Hoșgo¨r 2011). However, according to Goode’s hypothesis, when divorce
becomes more common, lower educated women will also experience divorce and
eventually,theeffectofeducationwillbereversed,withlowereducatedbeingmore
likelytodivorce(DeGraafandKalmijn2006).Wethereforeanticipatethatwomen
withahighereducationwillhaveahigherriskofdivorceintheearlieryearsofour
observation period, when divorce was relatively uncommon (Hypothesis 5).
Additionally, we expect thatthe effect of women’s education on divorce will be
shapedbytheregionalcontext.Similarly,weexpectthatinregionswheredivorceis
a rare phenomenon, the probability of divorce is higher for women with higher
education, relative to women with less education (Hypothesis 6). Additionally, in
less metropolitan, rural areas, where divorce is uncommon and less accepted, the
probability of divorce is higher for women with higher education compared to
women with less education (Hypothesis 7). With respect to wealthier regions, we
hypothesize that the relation between the economic context and the probability of
divorce will be most pronounced amongst those with less education; that is, better
socioeconomiccircumstancesstabilizemarriagesamongstthosewithlesseducation
and the divorce risk of higher educated women will increase relative to those of
lower educated women (Hypothesis 8).
123
616 K.Caarls,H.A.G.deValk
5 Micro-level Indicators of Divorce
The main focus of our study is on regional variance in divorce and changes over
time. In order to do so we, however, control for a range of micro-level
characteristics. We include indicators that have proved to be strong predictors of
divorce in a range of earlier studies (for review articles, see Amato 2000, 2010;
AmatoandJames2010;LyngstadandJalovaara2010;Ha¨rko¨nen2013).Inlinewith
these previous works, we control for union (like the duration of marriage, age at
marriage, age heterogamy, children born in the union, children from other than
spouse) and individual characteristics (like childhood place of residence, mother’s
literacy).IntheTurkishcontext,thereareseveralunion-specificcharacteristicsthat
also are essential to include (arranged marriages, consanguineous marriages) in
addition to a distinction between the different ethnic groups in the country
(indicated by mother tongue Kurdish) (Kavas 2010; Yavuz 2008). In line with the
hypotheses formulated before, we are also particularly interested in the interaction
between individual educational attainment (micro) and the regional diversity and
cohort and period changes (macro). Educational attainment of the women is a key
micro-level variable.
6 Data and Method
The data used in our analyses are the Turkish Demographic and Health Survey
(DHS),waves2003and2008.Inthesesurveys,householdswererandomlysampled
within the 12 regions of Turkey. In each of these households, all women that were
present in the household, or who usually live in that household, have been
interviewediftheywerebetweentheagesof15and49andhadbeenevermarried.
FormoredetailsabouttheDHSsurveys,theirtargetpopulationandresponserates,
please see the full DHS 2003 and DHS 2008 reports (DHS 2004: p. 183, 2009:
p. 222).
Two waves of data collection have been pooled, providing us with a robust
number of divorced and married Turkish women (742 and 15,480, respectively),1
covering marriages that took place between 1967 and 2008. The survey contains a
wide range of demographic andhealth-based questions, andit includesahistoryof
women’smarriages.Asthenumberofrecordedmarriagesvariesperwave,andthe
number of women with more than one marriage is limited, we will focus on first
marriagesonly.2Weconsiderwhetherthesefirstmarriagesendedindivorceornot.
1 Since we pool the two surveys, we de-normalized the weights using the following formula: new
weight=(DHSsampleweight/1,000,000)*(femalepopulationaged15–49insurveyyear/samplesizeof
survey year). Population numbers were derived from the 2000 and 2005 General Population Census
(United Nations (UN) 2015). These new weights were added to the data file and used for descriptive
analysesonly.
2 Ouranalyticalsampleofdivorcedwomenincludedboththosewhodidnotaswellaswhodidremarry
althoughthelattercategorywasverylimited:in2003,only3.0%(n=244)womenandin2008only
2.5%(n=186)remarried.
123
RegionalDiffusionofDivorceinTurkey 617
Using the retrospective information, we constructed a person-period file with
respondents’informationonayearlybasis.Wefollowedrespondentsfromtheyear
oftheirfirstmarriageuntildivorceorincaseofcensoringbythetimeofthesurvey
or by the death of the spouse. Additionally, we excluded respondents from whom
we did not have complete information concerning the start and end years of their
marriage(n=15andn=23,respectively).Thisresultedinadatasetconsistingof
15,418 respondents; 726 respondents experienced divorce or separation. The first
divorce occurred in 1973 and the last occurred in 2008. Our definition of divorce
includes women who are living together in an unmarried or a married union as the
Turkish DHS survey does not distinguish between them. Although we might
thereforeslightlyoverestimatethenumberofmarriedwomen,weexpecttheextent
of this bias to be minimal, as non-marital cohabitation hardly occurs in Turkey
(Yavuz 2008).
Since we want to explore the effects of individual and context-level factors on
women’s probability of divorce, we use multilevel discrete-time logistic regression
models that enable us to simultaneously use explanatory variables at these two
levels (i.e. individual and regional) (Snijders and Bosker 1999). The person-period
fileconsistedof222,616person-yearsandwedifferentiatedbetween12regions.We
assessed the duration dependency by using the number of years of marriage. We
tested for nonlinear effects, and the inclusion of a linear and a squared term fitted
thedatabest.Thetime-varyingvariableswerelaggedwithoneyear,whichisinline
with standard event history procedures (Singer and Willett 2003).
We first modelled a null model, which includes the random intercept and the
variable for duration of marriage (cf. Hox 2002: p. 81). To account for the
hierarchical structure of the data, all subsequent models include the random
intercept. Next, we includedall individual variables. In oursubsequent models, we
include the regional variables (due to our limited sample size at the regional level
and to avoid multicollinearity, we decided to estimate our regional variables
separately), and we additionally estimate models to examine period and marriage
cohort effects. Finally, we investigate interaction effects to examine whether the
effect of education has changed over time and whether the regional characteristics
shape the effect of education.
6.1 Measures of Contextual Variables
We distinguished 12 geographical regions (NUTS-1) as the region in which
respondents living at the time of survey. Three variables were constructed on the
regionallevel.First,wewereinterestedinregional CDR.Thisinformationwasnot
availablefortheentiretimeperiodwewereinterestedin.Wewere,however,ableto
obtain information about divorces per province. Today, Turkey is divided in 81
provinces(before1989,therewere67provinces,butseveralchangesbetween1989
and 1999 resulted in 81 provinces since then [for detailed information about
Turkey’s administrative divisions, see statoids.com/utr.html)]. Provincial crude
divorce rates were available through marriage and divorce statistics from TurkStat
[formally State Institute for Statistics (SIS)] for the period 2001–2008. For the
period 1970–2000, the SIS provided only information about the total number of
123
618 K.Caarls,H.A.G.deValk
divorces per province. Using the six censuses that were carried out by the State
Institute for Statistics from 1970–2000, we imputed the missing years to arrive at
populationestimatesforeachyear,foreachprovince.Crudedivorcerateswerethen
computed using the number of divorces and population estimates by province.
Usingtheseprovincialcrudedivorcerates,wecalculatedthecrudedivorcerate for
each ofthe 12regions bytaking the(weighted bypopulation size) average divorce
rate of the provinces in each region for each year (see Fig. 1, discussed below).
Second, using information about population density (population per km2) per
region for the period 2001–2009 (OECD 2015) as a proxy for urbanization, we
classified the 12 regions as rural (0–65), intermediate (66–99), and urban (100[).
As this information was only available for a limited time period, we explored the
variationoverthisdistributiontime,whichappearedalmostconstantforallregions
(seeFig.2,discussedbelow).Therefore,wedecidedtoincluderegionalpopulation
density as a time-constant variable.
ThethirdcontextualvariableistheaverageshareoftheregionalGDPpercapita
withinthetotalcountry-levelGDPpercapitafortheperiod1987–2000.Information
concerning Turkish GDP on regional or provincial level was also only available
from SIS for a limited period (1987–2001). Using population figures for each
region, we first calculated regional GDP per capita (regional GDP/regional
population) for the period 1987–2000. When examining the regional share of the
total country-level GDP per capita for this time period, our analyses demonstrated
littleregional variationovertimedespiteanoverallincrease inGDP(figures avail-
ableuponrequest).Thisledustoincludeatime-constantvariablethatcapturesthe
average share of the regional GDP per capita as a percentage of the total country-
level GDP per capita.
Fig.1 Crudedivorceratesbyprovince,1970–2011Source:Authors’calculations(basedon:Turkstat
2001–2009;SIS1970–2000)
123